Add a variant of compartment call benchmark that actually uses some stack.

This is to check the effect of the stack low water mark.
Also tests to make sure the stack is being properly cleared.

Author: ronorton

benchmarks/stack-usage/callee.cc

@@ -0,0 +1,15 @@
+#include "callee.h"
+#include "../timing.h"
+
+int  noop_return(size_t s) 
+{ 
+	use_stack(s);
+	return rdcycle(); 
+}
+
+int  noop_call(int start)
+{
+	int end = rdcycle();
+	check_stack_zeroed();
+	return end - start;
+}

benchmarks/stack-usage/callee.h

@@ -0,0 +1,4 @@
+#include <compartment.h>
+
+int __cheri_compartment("callee") noop_return(size_t s);
+int __cheri_compartment("callee") noop_call(int start);

benchmarks/stack-usage/caller.cc

@@ -0,0 +1,22 @@
+#include "../timing.h"
+#include "callee.h"
+#include <compartment.h>
+#include <locks.hh>
+
+void __cheri_compartment("caller") run()
+{
+	MessageBuilder<ImplicitUARTOutput> out;
+	out.format("#board\tstack_use\tcall\treturn\n");
+	for (size_t s = 1; s < 0x1000; s <<= 1)
+	{
+		auto [callPath, returnPath] = CHERI::with_interrupts_disabled([&]() {
+			use_stack(s);
+			auto callPath = noop_call(rdcycle());
+			auto returnPath = noop_return(s);
+			returnPath = rdcycle() - returnPath;
+			check_stack_zeroed();
+			return std::tuple{callPath, returnPath};
+		});
+		out.format(__XSTRING(BOARD) "\t{}\t{}\t{}\n", s, callPath, returnPath);
+	}
+}

benchmarks/stack-usage/xmake.lua

@@ -0,0 +1,39 @@
+-- Copyright Microsoft and CHERIoT Contributors.
+-- SPDX-License-Identifier: MIT
+
+set_project("CHERIoT stack-usage benchmark");
+sdkdir = "../../sdk"
+includes(sdkdir)
+set_toolchains("cheriot-clang")
+
+-- Support libraries
+includes(path.join(sdkdir, "lib/freestanding"),
+         path.join(sdkdir, "lib/atomic"),
+         path.join(sdkdir, "lib/crt"))
+
+option("board")
+    set_default("sail")
+
+compartment("callee")
+    add_files("callee.cc")
+
+compartment("caller")
+    add_defines("BOARD=" .. tostring(get_config("board")))
+    add_files("caller.cc")
+
+-- Firmware image for the example.
+firmware("stack-usage-benchmark")
+    add_deps("crt", "freestanding", "atomic")
+    add_deps("caller", "callee")
+    on_load(function(target)
+        target:values_set("board", "$(board)")
+        target:values_set("threads", {
+            {
+                compartment = "caller",
+                priority = 1,
+                entry_point = "run",
+                stack_size = 0x1000,
+                trusted_stack_frames = 3
+            }
+        }, {expand = false})
+    end)

benchmarks/timing.h

@@ -36,4 +36,28 @@ namespace
 		CHERI::Capability stack{cspRegister};
 		return stack.length();
 	}
+
+	void use_stack(size_t s)
+	{
+		volatile uint8_t stack_array[s];
+		stack_array[0] = 1;
+	}
+
+	void check_stack_zeroed()
+	{
+		register void **cspRegister asm("csp");
+		asm("" : "=C"(cspRegister));
+		CHERI::Capability stack{cspRegister};
+		CHERI::Capability<void*> stackP{stack};
+		stackP.address() = stack.base();
+		while(stackP.address() < stack.address())
+		{
+			CHERI::Capability<void> value{*stackP};
+			if (value != NULL)
+			{
+				__builtin_trap();
+			}
+			stackP += 1;
+		}
+	}
 } // namespace